Why no fold and unfold for 5d and 3d tensors?

2

This post uses the more flexible tensor.unfold method to compute the convolution and you could use it as a template for a 1d or 3d conv layer.